Posts

Status Update 12

This Week
I pushed our project onto Heroku so that it is publicly available at tensai.herokuapp.com. I also recorded myself playing the games and using practice mode for use in the team video, which Dakota edited. In addition to taking part in a few user studies, I also slightly modified the evaluation code so that it only uses the evaluation function for problematic characters if the recognizer indicates it is a problematic character (i.e. 三、王、玉、正).

Next Week
Prepare for final presentation and edit final report to improve flow.

Status Update 11

This Week
Short week due to Thanksgiving Break. Wrote about implementation of server and character recognition in final report. Fixed Japanese hovertext in the about page and added details discussing the feedback provided in practice mode. Developed unit tests for server functionality and evaluation code.

Next Week
Conduct user studies and summarize results in final report, then make changes based on user feedback as feasible. Edit final report for flow and readability

Status Update 10

This Week Over the weekend we successfully connected the server sketch recognition to both the ninja star game and the box stacking game, testing their behavior to make sure that the recognizer worked with a reasonable amount of accuracy. I also added Japanese hovertext to the menu buttons, researching how to translate the button names properly. In order to minimize collisions between similar looking characters in different character sets (e.g. katakana カ vs kanji 力), I made distinct recognizers for each set of characters, and reconfigured the server recognition functions to use the correct recognizer for a given input.
Next Week I am currently in the process of adding unit tests for server code and some $P recognizer functions. I also found a few similar-looking kanji which the recognizer often confuses, so I will work on modifying the recognition code to manually check those characters should they come up (they can be distinguished by number of strokes). Additionally, we will need to…

Status Update 9

This Week
Added to feedback provided in handwriting evaluation by coloring incorrect strokes based on the type of error that was made. An orange stroke indicates that the stroke was written in the wrong direction, a blue stroke indicates that the stroke was written out of order, and a red stroke indicates that the stroke is incorrect. If the character is not visually correct (i.e. identified by the $P recognizer), then the whole character is marked wrong. Additionally, I arranged the pages so that the opening page is the home menu, which then navigates to the specific usage modes.

Next Week
Next week we will present the current status of our work at SRLx. Additionally I will work on checking the recognition of kanji and work on getting our project ready for user studies.

Status Update 8

This Week
We finally met with Dr. Waugh and received her feedback on how students learn Japanese characters and the kind of mistakes they make in handwriting. In addition to working on the prior works section of the report, I also worked on evaluating user input sketches stroke by stroke. It currently recognizes ever even-numbered stroke in reverse order, so that will need to be fixed. I also created sketch data for the grade 1 Kyoiku kanji and will test them using the $P recognizer and the more naive stroke matching algorithm I worked on.

Next Week
Once the sketchpad code is integrated into the ninja star game, it can be connected to the server and tested to see if characters can be recognized in a satisfactory manner. With work on the training menus mostly done, the sketchpad code can also be added to provide visible feedback to users.

Status Update 7

This Week
Focused working on the back-end, first setting up communication between the server and client through Socket.io. I then fixed the update delay in the undo button, and modified its code so that it removes the most recently added stroke from sketchData. I also added the $P recognizer code to the server, and created data sets for both hiragana and katakana, testing them to ensure that the recognizer produced the expected results.

Next Week
I scheduled a meeting with Dr. Waugh from the Japanese department, though she was unable to attend, so next week I hope to schedule the meeting so that we can receive her feedback. Additional tasks include the prior works section of the report, adding additional character data sets to increase consistency, and adding algorithms to catch errors in stroke order and direction. Another direction to pursue this upcoming week is to integrate the sketchpad code with the ninja star game and begin testing that behavior.